Use a precompiled eGPU driving model (#38930)

* Ship precompiled eGPU model and camera warps

Compile f78ed37d-afad-4dbc-8050-40ea885eedde/12864 through xx/ml_tools/openpilot_compile using the pinned tinygrad version.

* Precompile the existing master driving model

Use the unchanged master ONNX (SHA-256 6fee5937923c74848df4a63f6239eb6331c6274dd4bdb7a5d6ec0388a8b543d5) instead of updating the trained model.

* Compile camera warps on device

* Remove obsolete ONNX chunking and big model build check

* Chunk model artifacts only during release packaging

* Require model and camera warps for Chestnut readiness

* Recompile precompiled CPU helpers for the runtime host

* Ship the eGPU model with an ARM submission helper

* Exempt model pickles from the build product size limit

SConstruct

@@ -342,6 +342,8 @@ AddPostAction(BUILD_TARGETS or [Dir('.')], prune_cache_dir)
 def check_build_product_size(target, source, env):
   limit = 50 * 1024 * 1024  # GitHub max size
   for t in target:
+    if str(t).endswith('.pkl'):  # chunked during release packaging
+      continue
     if hasattr(t, 'isfile') and t.isfile() and (size := os.path.getsize(t.abspath)) > limit:
       raise SCons.Errors.UserError(f"{t} is {size / (1024 * 1024):.1f} MiB, exceeding the {limit / (1024 * 1024):.1f} MiB limit")
 if not GetOption('extras'):